When it's time to pick, everyone types what they're in the mood for, in their own words. Claude searches the club's library for each person's mood, then judges every candidate against every member as a strong match, a partial match or a miss. The app offers three picks, we rank them, and once a book wins, it suggests the days most of us are free and sends the calendar invite.

Our second book sounded like it fit all of us, and we hated it. It was just a bad book, but it sent me to the research on group recommenders, which is about turning several people's tastes into one choice. There I found a separate problem: the app was averaging. A strong match scored 1, a partial 0.5 and a miss 0, so a book that was strong for three of us and a miss for the fourth scored 0.75, the same as one that was strong for two and partial for two. So the average hid the person who'd hate it.

The fix is called Average without Misery. It comes from a study where people chose TV for a group and many quietly took misery into account (Masthoff, 2004). It's the same average with one rule added: any book that misses anyone is out. If every book misses someone, the app falls back to the books with the fewest misses and tells us. I kept the average ranking the rest, because close groups tend to lean toward whatever pleases the most people (Gartrell et al., 2010).

Instead of taking the top three scores, the app picks the three books as a set, so as many of us as possible get one we're excited about. If one person's mood is unlike everyone else's, one slot can go to a book for them.